source: branches/samba-3.5.x/source3/lib/dbwrap_ctdb.c

Last change on this file was 429, checked in by Silvan Scherrer, 15 years ago

Samba 3.5.x: trunk update to 3.5.2

File size: 33.5 KB
Line 
1/*
2 Unix SMB/CIFS implementation.
3 Database interface wrapper around ctdbd
4 Copyright (C) Volker Lendecke 2007-2009
5 Copyright (C) Michael Adam 2009
6
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
11
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
16
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>.
19*/
20
21#include "includes.h"
22#ifdef CLUSTER_SUPPORT
23#include "ctdb.h"
24#include "ctdb_private.h"
25#include "ctdbd_conn.h"
26#include "g_lock.h"
27
28struct db_ctdb_transaction_handle {
29 struct db_ctdb_ctx *ctx;
30 /*
31 * we store the reads and writes done under a transaction:
32 * - one list stores both reads and writes (m_all),
33 * - the other just writes (m_write)
34 */
35 struct ctdb_marshall_buffer *m_all;
36 struct ctdb_marshall_buffer *m_write;
37 uint32_t nesting;
38 bool nested_cancel;
39 char *lock_name;
40};
41
42struct db_ctdb_ctx {
43 struct db_context *db;
44 struct tdb_wrap *wtdb;
45 uint32 db_id;
46 struct db_ctdb_transaction_handle *transaction;
47 struct g_lock_ctx *lock_ctx;
48};
49
50struct db_ctdb_rec {
51 struct db_ctdb_ctx *ctdb_ctx;
52 struct ctdb_ltdb_header header;
53};
54
55static NTSTATUS tdb_error_to_ntstatus(struct tdb_context *tdb)
56{
57 NTSTATUS status;
58 enum TDB_ERROR tret = tdb_error(tdb);
59
60 switch (tret) {
61 case TDB_ERR_EXISTS:
62 status = NT_STATUS_OBJECT_NAME_COLLISION;
63 break;
64 case TDB_ERR_NOEXIST:
65 status = NT_STATUS_OBJECT_NAME_NOT_FOUND;
66 break;
67 default:
68 status = NT_STATUS_INTERNAL_DB_CORRUPTION;
69 break;
70 }
71
72 return status;
73}
74
75
76/**
77 * fetch a record from the tdb, separating out the header
78 * information and returning the body of the record.
79 */
80static NTSTATUS db_ctdb_ltdb_fetch(struct db_ctdb_ctx *db,
81 TDB_DATA key,
82 struct ctdb_ltdb_header *header,
83 TALLOC_CTX *mem_ctx,
84 TDB_DATA *data)
85{
86 TDB_DATA rec;
87 NTSTATUS status;
88
89 rec = tdb_fetch(db->wtdb->tdb, key);
90 if (rec.dsize < sizeof(struct ctdb_ltdb_header)) {
91 status = NT_STATUS_NOT_FOUND;
92 if (data) {
93 ZERO_STRUCTP(data);
94 }
95 if (header) {
96 header->dmaster = (uint32_t)-1;
97 header->rsn = 0;
98 }
99 goto done;
100 }
101
102 if (header) {
103 *header = *(struct ctdb_ltdb_header *)rec.dptr;
104 }
105
106 if (data) {
107 data->dsize = rec.dsize - sizeof(struct ctdb_ltdb_header);
108 if (data->dsize == 0) {
109 data->dptr = NULL;
110 } else {
111 data->dptr = (unsigned char *)talloc_memdup(mem_ctx,
112 rec.dptr
113 + sizeof(struct ctdb_ltdb_header),
114 data->dsize);
115 if (data->dptr == NULL) {
116 status = NT_STATUS_NO_MEMORY;
117 goto done;
118 }
119 }
120 }
121
122 status = NT_STATUS_OK;
123
124done:
125 SAFE_FREE(rec.dptr);
126 return status;
127}
128
129/*
130 * Store a record together with the ctdb record header
131 * in the local copy of the database.
132 */
133static NTSTATUS db_ctdb_ltdb_store(struct db_ctdb_ctx *db,
134 TDB_DATA key,
135 struct ctdb_ltdb_header *header,
136 TDB_DATA data)
137{
138 TALLOC_CTX *tmp_ctx = talloc_stackframe();
139 TDB_DATA rec;
140 int ret;
141
142 rec.dsize = data.dsize + sizeof(struct ctdb_ltdb_header);
143 rec.dptr = (uint8_t *)talloc_size(tmp_ctx, rec.dsize);
144
145 if (rec.dptr == NULL) {
146 talloc_free(tmp_ctx);
147 return NT_STATUS_NO_MEMORY;
148 }
149
150 memcpy(rec.dptr, header, sizeof(struct ctdb_ltdb_header));
151 memcpy(sizeof(struct ctdb_ltdb_header) + (uint8_t *)rec.dptr, data.dptr, data.dsize);
152
153 ret = tdb_store(db->wtdb->tdb, key, rec, TDB_REPLACE);
154
155 talloc_free(tmp_ctx);
156
157 return (ret == 0) ? NT_STATUS_OK
158 : tdb_error_to_ntstatus(db->wtdb->tdb);
159
160}
161
162/*
163 form a ctdb_rec_data record from a key/data pair
164
165 note that header may be NULL. If not NULL then it is included in the data portion
166 of the record
167 */
168static struct ctdb_rec_data *db_ctdb_marshall_record(TALLOC_CTX *mem_ctx, uint32_t reqid,
169 TDB_DATA key,
170 struct ctdb_ltdb_header *header,
171 TDB_DATA data)
172{
173 size_t length;
174 struct ctdb_rec_data *d;
175
176 length = offsetof(struct ctdb_rec_data, data) + key.dsize +
177 data.dsize + (header?sizeof(*header):0);
178 d = (struct ctdb_rec_data *)talloc_size(mem_ctx, length);
179 if (d == NULL) {
180 return NULL;
181 }
182 d->length = length;
183 d->reqid = reqid;
184 d->keylen = key.dsize;
185 memcpy(&d->data[0], key.dptr, key.dsize);
186 if (header) {
187 d->datalen = data.dsize + sizeof(*header);
188 memcpy(&d->data[key.dsize], header, sizeof(*header));
189 memcpy(&d->data[key.dsize+sizeof(*header)], data.dptr, data.dsize);
190 } else {
191 d->datalen = data.dsize;
192 memcpy(&d->data[key.dsize], data.dptr, data.dsize);
193 }
194 return d;
195}
196
197
198/* helper function for marshalling multiple records */
199static struct ctdb_marshall_buffer *db_ctdb_marshall_add(TALLOC_CTX *mem_ctx,
200 struct ctdb_marshall_buffer *m,
201 uint64_t db_id,
202 uint32_t reqid,
203 TDB_DATA key,
204 struct ctdb_ltdb_header *header,
205 TDB_DATA data)
206{
207 struct ctdb_rec_data *r;
208 size_t m_size, r_size;
209 struct ctdb_marshall_buffer *m2 = NULL;
210
211 r = db_ctdb_marshall_record(talloc_tos(), reqid, key, header, data);
212 if (r == NULL) {
213 talloc_free(m);
214 return NULL;
215 }
216
217 if (m == NULL) {
218 m = (struct ctdb_marshall_buffer *)talloc_zero_size(
219 mem_ctx, offsetof(struct ctdb_marshall_buffer, data));
220 if (m == NULL) {
221 goto done;
222 }
223 m->db_id = db_id;
224 }
225
226 m_size = talloc_get_size(m);
227 r_size = talloc_get_size(r);
228
229 m2 = (struct ctdb_marshall_buffer *)talloc_realloc_size(
230 mem_ctx, m, m_size + r_size);
231 if (m2 == NULL) {
232 talloc_free(m);
233 goto done;
234 }
235
236 memcpy(m_size + (uint8_t *)m2, r, r_size);
237
238 m2->count++;
239
240done:
241 talloc_free(r);
242 return m2;
243}
244
245/* we've finished marshalling, return a data blob with the marshalled records */
246static TDB_DATA db_ctdb_marshall_finish(struct ctdb_marshall_buffer *m)
247{
248 TDB_DATA data;
249 data.dptr = (uint8_t *)m;
250 data.dsize = talloc_get_size(m);
251 return data;
252}
253
254/*
255 loop over a marshalling buffer
256
257 - pass r==NULL to start
258 - loop the number of times indicated by m->count
259*/
260static struct ctdb_rec_data *db_ctdb_marshall_loop_next(struct ctdb_marshall_buffer *m, struct ctdb_rec_data *r,
261 uint32_t *reqid,
262 struct ctdb_ltdb_header *header,
263 TDB_DATA *key, TDB_DATA *data)
264{
265 if (r == NULL) {
266 r = (struct ctdb_rec_data *)&m->data[0];
267 } else {
268 r = (struct ctdb_rec_data *)(r->length + (uint8_t *)r);
269 }
270
271 if (reqid != NULL) {
272 *reqid = r->reqid;
273 }
274
275 if (key != NULL) {
276 key->dptr = &r->data[0];
277 key->dsize = r->keylen;
278 }
279 if (data != NULL) {
280 data->dptr = &r->data[r->keylen];
281 data->dsize = r->datalen;
282 if (header != NULL) {
283 data->dptr += sizeof(*header);
284 data->dsize -= sizeof(*header);
285 }
286 }
287
288 if (header != NULL) {
289 if (r->datalen < sizeof(*header)) {
290 return NULL;
291 }
292 *header = *(struct ctdb_ltdb_header *)&r->data[r->keylen];
293 }
294
295 return r;
296}
297
298/**
299 * CTDB transaction destructor
300 */
301static int db_ctdb_transaction_destructor(struct db_ctdb_transaction_handle *h)
302{
303 NTSTATUS status;
304
305 status = g_lock_unlock(h->ctx->lock_ctx, h->lock_name);
306 if (!NT_STATUS_IS_OK(status)) {
307 DEBUG(0, ("g_lock_unlock failed: %s\n", nt_errstr(status)));
308 return -1;
309 }
310 return 0;
311}
312
313/**
314 * CTDB dbwrap API: transaction_start function
315 * starts a transaction on a persistent database
316 */
317static int db_ctdb_transaction_start(struct db_context *db)
318{
319 struct db_ctdb_transaction_handle *h;
320 NTSTATUS status;
321 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
322 struct db_ctdb_ctx);
323
324 if (!db->persistent) {
325 DEBUG(0,("transactions not supported on non-persistent database 0x%08x\n",
326 ctx->db_id));
327 return -1;
328 }
329
330 if (ctx->transaction) {
331 ctx->transaction->nesting++;
332 return 0;
333 }
334
335 h = talloc_zero(db, struct db_ctdb_transaction_handle);
336 if (h == NULL) {
337 DEBUG(0,(__location__ " oom for transaction handle\n"));
338 return -1;
339 }
340
341 h->ctx = ctx;
342
343 h->lock_name = talloc_asprintf(h, "transaction_db_0x%08x",
344 (unsigned int)ctx->db_id);
345 if (h->lock_name == NULL) {
346 DEBUG(0, ("talloc_asprintf failed\n"));
347 TALLOC_FREE(h);
348 return -1;
349 }
350
351 /*
352 * Wait a day, i.e. forever...
353 */
354 status = g_lock_lock(ctx->lock_ctx, h->lock_name, G_LOCK_WRITE,
355 timeval_set(86400, 0));
356 if (!NT_STATUS_IS_OK(status)) {
357 DEBUG(0, ("g_lock_lock failed: %s\n", nt_errstr(status)));
358 TALLOC_FREE(h);
359 return -1;
360 }
361
362 talloc_set_destructor(h, db_ctdb_transaction_destructor);
363
364 ctx->transaction = h;
365
366 DEBUG(5,(__location__ " Started transaction on db 0x%08x\n", ctx->db_id));
367
368 return 0;
369}
370
371static bool pull_newest_from_marshall_buffer(struct ctdb_marshall_buffer *buf,
372 TDB_DATA key,
373 struct ctdb_ltdb_header *pheader,
374 TALLOC_CTX *mem_ctx,
375 TDB_DATA *pdata)
376{
377 struct ctdb_rec_data *rec = NULL;
378 struct ctdb_ltdb_header h;
379 bool found = false;
380 TDB_DATA data;
381 int i;
382
383 if (buf == NULL) {
384 return false;
385 }
386
387 ZERO_STRUCT(h);
388 ZERO_STRUCT(data);
389
390 /*
391 * Walk the list of records written during this
392 * transaction. If we want to read one we have already
393 * written, return the last written sample. Thus we do not do
394 * a "break;" for the first hit, this record might have been
395 * overwritten later.
396 */
397
398 for (i=0; i<buf->count; i++) {
399 TDB_DATA tkey, tdata;
400 uint32_t reqid;
401 struct ctdb_ltdb_header hdr;
402
403 ZERO_STRUCT(hdr);
404
405 rec = db_ctdb_marshall_loop_next(buf, rec, &reqid, &hdr, &tkey,
406 &tdata);
407 if (rec == NULL) {
408 return false;
409 }
410
411 if (tdb_data_equal(key, tkey)) {
412 found = true;
413 data = tdata;
414 h = hdr;
415 }
416 }
417
418 if (!found) {
419 return false;
420 }
421
422 if (pdata != NULL) {
423 data.dptr = (uint8_t *)talloc_memdup(mem_ctx, data.dptr,
424 data.dsize);
425 if ((data.dsize != 0) && (data.dptr == NULL)) {
426 return false;
427 }
428 *pdata = data;
429 }
430
431 if (pheader != NULL) {
432 *pheader = h;
433 }
434
435 return true;
436}
437
438/*
439 fetch a record inside a transaction
440 */
441static int db_ctdb_transaction_fetch(struct db_ctdb_ctx *db,
442 TALLOC_CTX *mem_ctx,
443 TDB_DATA key, TDB_DATA *data)
444{
445 struct db_ctdb_transaction_handle *h = db->transaction;
446 NTSTATUS status;
447 bool found;
448
449 found = pull_newest_from_marshall_buffer(h->m_write, key, NULL,
450 mem_ctx, data);
451 if (found) {
452 return 0;
453 }
454
455 status = db_ctdb_ltdb_fetch(h->ctx, key, NULL, mem_ctx, data);
456
457 if (NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND)) {
458 *data = tdb_null;
459 } else if (!NT_STATUS_IS_OK(status)) {
460 return -1;
461 }
462
463 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 1, key,
464 NULL, *data);
465 if (h->m_all == NULL) {
466 DEBUG(0,(__location__ " Failed to add to marshalling "
467 "record\n"));
468 data->dsize = 0;
469 talloc_free(data->dptr);
470 return -1;
471 }
472
473 return 0;
474}
475
476
477static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag);
478static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec);
479
480static struct db_record *db_ctdb_fetch_locked_transaction(struct db_ctdb_ctx *ctx,
481 TALLOC_CTX *mem_ctx,
482 TDB_DATA key)
483{
484 struct db_record *result;
485 TDB_DATA ctdb_data;
486
487 if (!(result = talloc(mem_ctx, struct db_record))) {
488 DEBUG(0, ("talloc failed\n"));
489 return NULL;
490 }
491
492 result->private_data = ctx->transaction;
493
494 result->key.dsize = key.dsize;
495 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
496 if (result->key.dptr == NULL) {
497 DEBUG(0, ("talloc failed\n"));
498 TALLOC_FREE(result);
499 return NULL;
500 }
501
502 result->store = db_ctdb_store_transaction;
503 result->delete_rec = db_ctdb_delete_transaction;
504
505 if (pull_newest_from_marshall_buffer(ctx->transaction->m_write, key,
506 NULL, result, &result->value)) {
507 return result;
508 }
509
510 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
511 if (ctdb_data.dptr == NULL) {
512 /* create the record */
513 result->value = tdb_null;
514 return result;
515 }
516
517 result->value.dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
518 result->value.dptr = NULL;
519
520 if ((result->value.dsize != 0)
521 && !(result->value.dptr = (uint8 *)talloc_memdup(
522 result, ctdb_data.dptr + sizeof(struct ctdb_ltdb_header),
523 result->value.dsize))) {
524 DEBUG(0, ("talloc failed\n"));
525 TALLOC_FREE(result);
526 }
527
528 SAFE_FREE(ctdb_data.dptr);
529
530 return result;
531}
532
533static int db_ctdb_record_destructor(struct db_record **recp)
534{
535 struct db_record *rec = talloc_get_type_abort(*recp, struct db_record);
536 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
537 rec->private_data, struct db_ctdb_transaction_handle);
538 int ret = h->ctx->db->transaction_commit(h->ctx->db);
539 if (ret != 0) {
540 DEBUG(0,(__location__ " transaction_commit failed\n"));
541 }
542 return 0;
543}
544
545/*
546 auto-create a transaction for persistent databases
547 */
548static struct db_record *db_ctdb_fetch_locked_persistent(struct db_ctdb_ctx *ctx,
549 TALLOC_CTX *mem_ctx,
550 TDB_DATA key)
551{
552 int res;
553 struct db_record *rec, **recp;
554
555 res = db_ctdb_transaction_start(ctx->db);
556 if (res == -1) {
557 return NULL;
558 }
559
560 rec = db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
561 if (rec == NULL) {
562 ctx->db->transaction_cancel(ctx->db);
563 return NULL;
564 }
565
566 /* destroy this transaction when we release the lock */
567 recp = talloc(rec, struct db_record *);
568 if (recp == NULL) {
569 ctx->db->transaction_cancel(ctx->db);
570 talloc_free(rec);
571 return NULL;
572 }
573 *recp = rec;
574 talloc_set_destructor(recp, db_ctdb_record_destructor);
575 return rec;
576}
577
578
579/*
580 stores a record inside a transaction
581 */
582static NTSTATUS db_ctdb_transaction_store(struct db_ctdb_transaction_handle *h,
583 TDB_DATA key, TDB_DATA data)
584{
585 TALLOC_CTX *tmp_ctx = talloc_new(h);
586 TDB_DATA rec;
587 struct ctdb_ltdb_header header;
588
589 ZERO_STRUCT(header);
590
591 /* we need the header so we can update the RSN */
592
593 if (!pull_newest_from_marshall_buffer(h->m_write, key, &header,
594 NULL, NULL)) {
595
596 rec = tdb_fetch(h->ctx->wtdb->tdb, key);
597
598 if (rec.dptr != NULL) {
599 memcpy(&header, rec.dptr,
600 sizeof(struct ctdb_ltdb_header));
601 rec.dsize -= sizeof(struct ctdb_ltdb_header);
602
603 /*
604 * a special case, we are writing the same
605 * data that is there now
606 */
607 if (data.dsize == rec.dsize &&
608 memcmp(data.dptr,
609 rec.dptr + sizeof(struct ctdb_ltdb_header),
610 data.dsize) == 0) {
611 SAFE_FREE(rec.dptr);
612 talloc_free(tmp_ctx);
613 return NT_STATUS_OK;
614 }
615 }
616 SAFE_FREE(rec.dptr);
617 }
618
619 header.dmaster = get_my_vnn();
620 header.rsn++;
621
622 h->m_all = db_ctdb_marshall_add(h, h->m_all, h->ctx->db_id, 0, key,
623 NULL, data);
624 if (h->m_all == NULL) {
625 DEBUG(0,(__location__ " Failed to add to marshalling "
626 "record\n"));
627 talloc_free(tmp_ctx);
628 return NT_STATUS_NO_MEMORY;
629 }
630
631 h->m_write = db_ctdb_marshall_add(h, h->m_write, h->ctx->db_id, 0, key, &header, data);
632 if (h->m_write == NULL) {
633 DEBUG(0,(__location__ " Failed to add to marshalling record\n"));
634 talloc_free(tmp_ctx);
635 return NT_STATUS_NO_MEMORY;
636 }
637
638 talloc_free(tmp_ctx);
639 return NT_STATUS_OK;
640}
641
642
643/*
644 a record store inside a transaction
645 */
646static NTSTATUS db_ctdb_store_transaction(struct db_record *rec, TDB_DATA data, int flag)
647{
648 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
649 rec->private_data, struct db_ctdb_transaction_handle);
650 NTSTATUS status;
651
652 status = db_ctdb_transaction_store(h, rec->key, data);
653 return status;
654}
655
656/*
657 a record delete inside a transaction
658 */
659static NTSTATUS db_ctdb_delete_transaction(struct db_record *rec)
660{
661 struct db_ctdb_transaction_handle *h = talloc_get_type_abort(
662 rec->private_data, struct db_ctdb_transaction_handle);
663 NTSTATUS status;
664
665 status = db_ctdb_transaction_store(h, rec->key, tdb_null);
666 return status;
667}
668
669/**
670 * Fetch the db sequence number of a persistent db directly from the db.
671 */
672static NTSTATUS db_ctdb_fetch_db_seqnum_from_db(struct db_ctdb_ctx *db,
673 uint64_t *seqnum)
674{
675 NTSTATUS status;
676 const char *keyname = CTDB_DB_SEQNUM_KEY;
677 TDB_DATA key;
678 TDB_DATA data;
679 struct ctdb_ltdb_header header;
680 TALLOC_CTX *mem_ctx = talloc_stackframe();
681
682 if (seqnum == NULL) {
683 return NT_STATUS_INVALID_PARAMETER;
684 }
685
686 key = string_term_tdb_data(keyname);
687
688 status = db_ctdb_ltdb_fetch(db, key, &header, mem_ctx, &data);
689 if (!NT_STATUS_IS_OK(status) &&
690 !NT_STATUS_EQUAL(status, NT_STATUS_NOT_FOUND))
691 {
692 goto done;
693 }
694
695 status = NT_STATUS_OK;
696
697 if (data.dsize != sizeof(uint64_t)) {
698 *seqnum = 0;
699 goto done;
700 }
701
702 *seqnum = *(uint64_t *)data.dptr;
703
704done:
705 TALLOC_FREE(mem_ctx);
706 return status;
707}
708
709/**
710 * Store the database sequence number inside a transaction.
711 */
712static NTSTATUS db_ctdb_store_db_seqnum(struct db_ctdb_transaction_handle *h,
713 uint64_t seqnum)
714{
715 NTSTATUS status;
716 const char *keyname = CTDB_DB_SEQNUM_KEY;
717 TDB_DATA key;
718 TDB_DATA data;
719
720 key = string_term_tdb_data(keyname);
721
722 data.dptr = (uint8_t *)&seqnum;
723 data.dsize = sizeof(uint64_t);
724
725 status = db_ctdb_transaction_store(h, key, data);
726
727 return status;
728}
729
730/*
731 commit a transaction
732 */
733static int db_ctdb_transaction_commit(struct db_context *db)
734{
735 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
736 struct db_ctdb_ctx);
737 NTSTATUS rets;
738 int status;
739 struct db_ctdb_transaction_handle *h = ctx->transaction;
740 uint64_t old_seqnum, new_seqnum;
741 int ret;
742
743 if (h == NULL) {
744 DEBUG(0,(__location__ " transaction commit with no open transaction on db 0x%08x\n", ctx->db_id));
745 return -1;
746 }
747
748 if (h->nested_cancel) {
749 db->transaction_cancel(db);
750 DEBUG(5,(__location__ " Failed transaction commit after nested cancel\n"));
751 return -1;
752 }
753
754 if (h->nesting != 0) {
755 h->nesting--;
756 return 0;
757 }
758
759 if (h->m_write == NULL) {
760 /*
761 * No changes were made, so don't change the seqnum,
762 * don't push to other node, just exit with success.
763 */
764 ret = 0;
765 goto done;
766 }
767
768 DEBUG(5,(__location__ " Commit transaction on db 0x%08x\n", ctx->db_id));
769
770 /*
771 * As the last db action before committing, bump the database sequence
772 * number. Note that this undoes all changes to the seqnum records
773 * performed under the transaction. This record is not meant to be
774 * modified by user interaction. It is for internal use only...
775 */
776 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &old_seqnum);
777 if (!NT_STATUS_IS_OK(rets)) {
778 DEBUG(1, (__location__ " failed to fetch the db sequence number "
779 "in transaction commit on db 0x%08x\n", ctx->db_id));
780 ret = -1;
781 goto done;
782 }
783
784 new_seqnum = old_seqnum + 1;
785
786 rets = db_ctdb_store_db_seqnum(h, new_seqnum);
787 if (!NT_STATUS_IS_OK(rets)) {
788 DEBUG(1, (__location__ "failed to store the db sequence number "
789 " in transaction commit on db 0x%08x\n", ctx->db_id));
790 ret = -1;
791 goto done;
792 }
793
794again:
795 /* tell ctdbd to commit to the other nodes */
796 rets = ctdbd_control_local(messaging_ctdbd_connection(),
797 CTDB_CONTROL_TRANS3_COMMIT,
798 h->ctx->db_id, 0,
799 db_ctdb_marshall_finish(h->m_write),
800 NULL, NULL, &status);
801 if (!NT_STATUS_IS_OK(rets) || status != 0) {
802 /*
803 * The TRANS3_COMMIT control should only possibly fail when a
804 * recovery has been running concurrently. In any case, the db
805 * will be the same on all nodes, either the new copy or the
806 * old copy. This can be detected by comparing the old and new
807 * local sequence numbers.
808 */
809 rets = db_ctdb_fetch_db_seqnum_from_db(ctx, &new_seqnum);
810 if (!NT_STATUS_IS_OK(rets)) {
811 DEBUG(1, (__location__ " failed to refetch db sequence "
812 "number after failed TRANS3_COMMIT\n"));
813 ret = -1;
814 goto done;
815 }
816
817 if (new_seqnum == old_seqnum) {
818 /* Recovery prevented all our changes: retry. */
819 goto again;
820 } else if (new_seqnum != (old_seqnum + 1)) {
821 DEBUG(0, (__location__ " ERROR: new_seqnum[%lu] != "
822 "old_seqnum[%lu] + (0 or 1) after failed "
823 "TRANS3_COMMIT - this should not happen!\n",
824 (unsigned long)new_seqnum,
825 (unsigned long)old_seqnum));
826 ret = -1;
827 goto done;
828 }
829 /*
830 * Recovery propagated our changes to all nodes, completing
831 * our commit for us - succeed.
832 */
833 }
834
835 ret = 0;
836
837done:
838 h->ctx->transaction = NULL;
839 talloc_free(h);
840 return ret;
841}
842
843
844/*
845 cancel a transaction
846 */
847static int db_ctdb_transaction_cancel(struct db_context *db)
848{
849 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
850 struct db_ctdb_ctx);
851 struct db_ctdb_transaction_handle *h = ctx->transaction;
852
853 if (h == NULL) {
854 DEBUG(0,(__location__ " transaction cancel with no open transaction on db 0x%08x\n", ctx->db_id));
855 return -1;
856 }
857
858 if (h->nesting != 0) {
859 h->nesting--;
860 h->nested_cancel = true;
861 return 0;
862 }
863
864 DEBUG(5,(__location__ " Cancel transaction on db 0x%08x\n", ctx->db_id));
865
866 ctx->transaction = NULL;
867 talloc_free(h);
868 return 0;
869}
870
871
872static NTSTATUS db_ctdb_store(struct db_record *rec, TDB_DATA data, int flag)
873{
874 struct db_ctdb_rec *crec = talloc_get_type_abort(
875 rec->private_data, struct db_ctdb_rec);
876
877 return db_ctdb_ltdb_store(crec->ctdb_ctx, rec->key, &(crec->header), data);
878}
879
880
881
882static NTSTATUS db_ctdb_delete(struct db_record *rec)
883{
884 TDB_DATA data;
885
886 /*
887 * We have to store the header with empty data. TODO: Fix the
888 * tdb-level cleanup
889 */
890
891 ZERO_STRUCT(data);
892
893 return db_ctdb_store(rec, data, 0);
894
895}
896
897static int db_ctdb_record_destr(struct db_record* data)
898{
899 struct db_ctdb_rec *crec = talloc_get_type_abort(
900 data->private_data, struct db_ctdb_rec);
901
902 DEBUG(10, (DEBUGLEVEL > 10
903 ? "Unlocking db %u key %s\n"
904 : "Unlocking db %u key %.20s\n",
905 (int)crec->ctdb_ctx->db_id,
906 hex_encode_talloc(data, (unsigned char *)data->key.dptr,
907 data->key.dsize)));
908
909 if (tdb_chainunlock(crec->ctdb_ctx->wtdb->tdb, data->key) != 0) {
910 DEBUG(0, ("tdb_chainunlock failed\n"));
911 return -1;
912 }
913
914 return 0;
915}
916
917static struct db_record *fetch_locked_internal(struct db_ctdb_ctx *ctx,
918 TALLOC_CTX *mem_ctx,
919 TDB_DATA key)
920{
921 struct db_record *result;
922 struct db_ctdb_rec *crec;
923 NTSTATUS status;
924 TDB_DATA ctdb_data;
925 int migrate_attempts = 0;
926
927 if (!(result = talloc(mem_ctx, struct db_record))) {
928 DEBUG(0, ("talloc failed\n"));
929 return NULL;
930 }
931
932 if (!(crec = TALLOC_ZERO_P(result, struct db_ctdb_rec))) {
933 DEBUG(0, ("talloc failed\n"));
934 TALLOC_FREE(result);
935 return NULL;
936 }
937
938 result->private_data = (void *)crec;
939 crec->ctdb_ctx = ctx;
940
941 result->key.dsize = key.dsize;
942 result->key.dptr = (uint8 *)talloc_memdup(result, key.dptr, key.dsize);
943 if (result->key.dptr == NULL) {
944 DEBUG(0, ("talloc failed\n"));
945 TALLOC_FREE(result);
946 return NULL;
947 }
948
949 /*
950 * Do a blocking lock on the record
951 */
952again:
953
954 if (DEBUGLEVEL >= 10) {
955 char *keystr = hex_encode_talloc(result, key.dptr, key.dsize);
956 DEBUG(10, (DEBUGLEVEL > 10
957 ? "Locking db %u key %s\n"
958 : "Locking db %u key %.20s\n",
959 (int)crec->ctdb_ctx->db_id, keystr));
960 TALLOC_FREE(keystr);
961 }
962
963 if (tdb_chainlock(ctx->wtdb->tdb, key) != 0) {
964 DEBUG(3, ("tdb_chainlock failed\n"));
965 TALLOC_FREE(result);
966 return NULL;
967 }
968
969 result->store = db_ctdb_store;
970 result->delete_rec = db_ctdb_delete;
971 talloc_set_destructor(result, db_ctdb_record_destr);
972
973 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
974
975 /*
976 * See if we have a valid record and we are the dmaster. If so, we can
977 * take the shortcut and just return it.
978 */
979
980 if ((ctdb_data.dptr == NULL) ||
981 (ctdb_data.dsize < sizeof(struct ctdb_ltdb_header)) ||
982 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster != get_my_vnn()
983#if 0
984 || (random() % 2 != 0)
985#endif
986) {
987 SAFE_FREE(ctdb_data.dptr);
988 tdb_chainunlock(ctx->wtdb->tdb, key);
989 talloc_set_destructor(result, NULL);
990
991 migrate_attempts += 1;
992
993 DEBUG(10, ("ctdb_data.dptr = %p, dmaster = %u (%u)\n",
994 ctdb_data.dptr, ctdb_data.dptr ?
995 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster : -1,
996 get_my_vnn()));
997
998 status = ctdbd_migrate(messaging_ctdbd_connection(),ctx->db_id, key);
999 if (!NT_STATUS_IS_OK(status)) {
1000 DEBUG(5, ("ctdb_migrate failed: %s\n",
1001 nt_errstr(status)));
1002 TALLOC_FREE(result);
1003 return NULL;
1004 }
1005 /* now its migrated, try again */
1006 goto again;
1007 }
1008
1009 if (migrate_attempts > 10) {
1010 DEBUG(0, ("db_ctdb_fetch_locked needed %d attempts\n",
1011 migrate_attempts));
1012 }
1013
1014 memcpy(&crec->header, ctdb_data.dptr, sizeof(crec->header));
1015
1016 result->value.dsize = ctdb_data.dsize - sizeof(crec->header);
1017 result->value.dptr = NULL;
1018
1019 if ((result->value.dsize != 0)
1020 && !(result->value.dptr = (uint8 *)talloc_memdup(
1021 result, ctdb_data.dptr + sizeof(crec->header),
1022 result->value.dsize))) {
1023 DEBUG(0, ("talloc failed\n"));
1024 TALLOC_FREE(result);
1025 }
1026
1027 SAFE_FREE(ctdb_data.dptr);
1028
1029 return result;
1030}
1031
1032static struct db_record *db_ctdb_fetch_locked(struct db_context *db,
1033 TALLOC_CTX *mem_ctx,
1034 TDB_DATA key)
1035{
1036 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1037 struct db_ctdb_ctx);
1038
1039 if (ctx->transaction != NULL) {
1040 return db_ctdb_fetch_locked_transaction(ctx, mem_ctx, key);
1041 }
1042
1043 if (db->persistent) {
1044 return db_ctdb_fetch_locked_persistent(ctx, mem_ctx, key);
1045 }
1046
1047 return fetch_locked_internal(ctx, mem_ctx, key);
1048}
1049
1050/*
1051 fetch (unlocked, no migration) operation on ctdb
1052 */
1053static int db_ctdb_fetch(struct db_context *db, TALLOC_CTX *mem_ctx,
1054 TDB_DATA key, TDB_DATA *data)
1055{
1056 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1057 struct db_ctdb_ctx);
1058 NTSTATUS status;
1059 TDB_DATA ctdb_data;
1060
1061 if (ctx->transaction) {
1062 return db_ctdb_transaction_fetch(ctx, mem_ctx, key, data);
1063 }
1064
1065 /* try a direct fetch */
1066 ctdb_data = tdb_fetch(ctx->wtdb->tdb, key);
1067
1068 /*
1069 * See if we have a valid record and we are the dmaster. If so, we can
1070 * take the shortcut and just return it.
1071 * we bypass the dmaster check for persistent databases
1072 */
1073 if ((ctdb_data.dptr != NULL) &&
1074 (ctdb_data.dsize >= sizeof(struct ctdb_ltdb_header)) &&
1075 (db->persistent ||
1076 ((struct ctdb_ltdb_header *)ctdb_data.dptr)->dmaster == get_my_vnn())) {
1077 /* we are the dmaster - avoid the ctdb protocol op */
1078
1079 data->dsize = ctdb_data.dsize - sizeof(struct ctdb_ltdb_header);
1080 if (data->dsize == 0) {
1081 SAFE_FREE(ctdb_data.dptr);
1082 data->dptr = NULL;
1083 return 0;
1084 }
1085
1086 data->dptr = (uint8 *)talloc_memdup(
1087 mem_ctx, ctdb_data.dptr+sizeof(struct ctdb_ltdb_header),
1088 data->dsize);
1089
1090 SAFE_FREE(ctdb_data.dptr);
1091
1092 if (data->dptr == NULL) {
1093 return -1;
1094 }
1095 return 0;
1096 }
1097
1098 SAFE_FREE(ctdb_data.dptr);
1099
1100 /* we weren't able to get it locally - ask ctdb to fetch it for us */
1101 status = ctdbd_fetch(messaging_ctdbd_connection(),ctx->db_id, key, mem_ctx, data);
1102 if (!NT_STATUS_IS_OK(status)) {
1103 DEBUG(5, ("ctdbd_fetch failed: %s\n", nt_errstr(status)));
1104 return -1;
1105 }
1106
1107 return 0;
1108}
1109
1110struct traverse_state {
1111 struct db_context *db;
1112 int (*fn)(struct db_record *rec, void *private_data);
1113 void *private_data;
1114};
1115
1116static void traverse_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1117{
1118 struct traverse_state *state = (struct traverse_state *)private_data;
1119 struct db_record *rec;
1120 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1121 /* we have to give them a locked record to prevent races */
1122 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, key);
1123 if (rec && rec->value.dsize > 0) {
1124 state->fn(rec, state->private_data);
1125 }
1126 talloc_free(tmp_ctx);
1127}
1128
1129static int traverse_persistent_callback(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1130 void *private_data)
1131{
1132 struct traverse_state *state = (struct traverse_state *)private_data;
1133 struct db_record *rec;
1134 TALLOC_CTX *tmp_ctx = talloc_new(state->db);
1135 int ret = 0;
1136 /* we have to give them a locked record to prevent races */
1137 rec = db_ctdb_fetch_locked(state->db, tmp_ctx, kbuf);
1138 if (rec && rec->value.dsize > 0) {
1139 ret = state->fn(rec, state->private_data);
1140 }
1141 talloc_free(tmp_ctx);
1142 return ret;
1143}
1144
1145static int db_ctdb_traverse(struct db_context *db,
1146 int (*fn)(struct db_record *rec,
1147 void *private_data),
1148 void *private_data)
1149{
1150 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1151 struct db_ctdb_ctx);
1152 struct traverse_state state;
1153
1154 state.db = db;
1155 state.fn = fn;
1156 state.private_data = private_data;
1157
1158 if (db->persistent) {
1159 /* for persistent databases we don't need to do a ctdb traverse,
1160 we can do a faster local traverse */
1161 return tdb_traverse(ctx->wtdb->tdb, traverse_persistent_callback, &state);
1162 }
1163
1164
1165 ctdbd_traverse(ctx->db_id, traverse_callback, &state);
1166 return 0;
1167}
1168
1169static NTSTATUS db_ctdb_store_deny(struct db_record *rec, TDB_DATA data, int flag)
1170{
1171 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1172}
1173
1174static NTSTATUS db_ctdb_delete_deny(struct db_record *rec)
1175{
1176 return NT_STATUS_MEDIA_WRITE_PROTECTED;
1177}
1178
1179static void traverse_read_callback(TDB_DATA key, TDB_DATA data, void *private_data)
1180{
1181 struct traverse_state *state = (struct traverse_state *)private_data;
1182 struct db_record rec;
1183 rec.key = key;
1184 rec.value = data;
1185 rec.store = db_ctdb_store_deny;
1186 rec.delete_rec = db_ctdb_delete_deny;
1187 rec.private_data = state->db;
1188 state->fn(&rec, state->private_data);
1189}
1190
1191static int traverse_persistent_callback_read(TDB_CONTEXT *tdb, TDB_DATA kbuf, TDB_DATA dbuf,
1192 void *private_data)
1193{
1194 struct traverse_state *state = (struct traverse_state *)private_data;
1195 struct db_record rec;
1196 rec.key = kbuf;
1197 rec.value = dbuf;
1198 rec.store = db_ctdb_store_deny;
1199 rec.delete_rec = db_ctdb_delete_deny;
1200 rec.private_data = state->db;
1201
1202 if (rec.value.dsize <= sizeof(struct ctdb_ltdb_header)) {
1203 /* a deleted record */
1204 return 0;
1205 }
1206 rec.value.dsize -= sizeof(struct ctdb_ltdb_header);
1207 rec.value.dptr += sizeof(struct ctdb_ltdb_header);
1208
1209 return state->fn(&rec, state->private_data);
1210}
1211
1212static int db_ctdb_traverse_read(struct db_context *db,
1213 int (*fn)(struct db_record *rec,
1214 void *private_data),
1215 void *private_data)
1216{
1217 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1218 struct db_ctdb_ctx);
1219 struct traverse_state state;
1220
1221 state.db = db;
1222 state.fn = fn;
1223 state.private_data = private_data;
1224
1225 if (db->persistent) {
1226 /* for persistent databases we don't need to do a ctdb traverse,
1227 we can do a faster local traverse */
1228 return tdb_traverse_read(ctx->wtdb->tdb, traverse_persistent_callback_read, &state);
1229 }
1230
1231 ctdbd_traverse(ctx->db_id, traverse_read_callback, &state);
1232 return 0;
1233}
1234
1235static int db_ctdb_get_seqnum(struct db_context *db)
1236{
1237 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1238 struct db_ctdb_ctx);
1239 return tdb_get_seqnum(ctx->wtdb->tdb);
1240}
1241
1242static int db_ctdb_get_flags(struct db_context *db)
1243{
1244 struct db_ctdb_ctx *ctx = talloc_get_type_abort(db->private_data,
1245 struct db_ctdb_ctx);
1246 return tdb_get_flags(ctx->wtdb->tdb);
1247}
1248
1249struct db_context *db_open_ctdb(TALLOC_CTX *mem_ctx,
1250 const char *name,
1251 int hash_size, int tdb_flags,
1252 int open_flags, mode_t mode)
1253{
1254 struct db_context *result;
1255 struct db_ctdb_ctx *db_ctdb;
1256 char *db_path;
1257 struct ctdbd_connection *conn;
1258
1259 if (!lp_clustering()) {
1260 DEBUG(10, ("Clustering disabled -- no ctdb\n"));
1261 return NULL;
1262 }
1263
1264 if (!(result = TALLOC_ZERO_P(mem_ctx, struct db_context))) {
1265 DEBUG(0, ("talloc failed\n"));
1266 TALLOC_FREE(result);
1267 return NULL;
1268 }
1269
1270 if (!(db_ctdb = TALLOC_P(result, struct db_ctdb_ctx))) {
1271 DEBUG(0, ("talloc failed\n"));
1272 TALLOC_FREE(result);
1273 return NULL;
1274 }
1275
1276 db_ctdb->transaction = NULL;
1277 db_ctdb->db = result;
1278
1279 conn = messaging_ctdbd_connection();
1280
1281 if (!NT_STATUS_IS_OK(ctdbd_db_attach(conn, name, &db_ctdb->db_id, tdb_flags))) {
1282 DEBUG(0, ("ctdbd_db_attach failed for %s\n", name));
1283 TALLOC_FREE(result);
1284 return NULL;
1285 }
1286
1287 db_path = ctdbd_dbpath(conn, db_ctdb, db_ctdb->db_id);
1288
1289 result->persistent = ((tdb_flags & TDB_CLEAR_IF_FIRST) == 0);
1290
1291 /* only pass through specific flags */
1292 tdb_flags &= TDB_SEQNUM;
1293
1294 /* honor permissions if user has specified O_CREAT */
1295 if (open_flags & O_CREAT) {
1296 chmod(db_path, mode);
1297 }
1298
1299 db_ctdb->wtdb = tdb_wrap_open(db_ctdb, db_path, hash_size, tdb_flags, O_RDWR, 0);
1300 if (db_ctdb->wtdb == NULL) {
1301 DEBUG(0, ("Could not open tdb %s: %s\n", db_path, strerror(errno)));
1302 TALLOC_FREE(result);
1303 return NULL;
1304 }
1305 talloc_free(db_path);
1306
1307 if (result->persistent) {
1308 db_ctdb->lock_ctx = g_lock_ctx_init(db_ctdb,
1309 ctdb_conn_msg_ctx(conn));
1310 if (db_ctdb->lock_ctx == NULL) {
1311 DEBUG(0, ("g_lock_ctx_init failed\n"));
1312 TALLOC_FREE(result);
1313 return NULL;
1314 }
1315 }
1316
1317 result->private_data = (void *)db_ctdb;
1318 result->fetch_locked = db_ctdb_fetch_locked;
1319 result->fetch = db_ctdb_fetch;
1320 result->traverse = db_ctdb_traverse;
1321 result->traverse_read = db_ctdb_traverse_read;
1322 result->get_seqnum = db_ctdb_get_seqnum;
1323 result->get_flags = db_ctdb_get_flags;
1324 result->transaction_start = db_ctdb_transaction_start;
1325 result->transaction_commit = db_ctdb_transaction_commit;
1326 result->transaction_cancel = db_ctdb_transaction_cancel;
1327
1328 DEBUG(3,("db_open_ctdb: opened database '%s' with dbid 0x%x\n",
1329 name, db_ctdb->db_id));
1330
1331 return result;
1332}
1333#endif
Note: See TracBrowser for help on using the repository browser.